How artificial intelligence can improve protein detection
(Nanowerk News) Small proteins play an important role in the regulation of immune responses, inflammation and neurodegenerative diseases. To better detect and study them, scientists at the Max-Planck-Institute for the Science of Light have combined one of the most effective microscopy methods, called iSCAT, with artificial intelligence.
Biological molecules such as proteins are major constituents of all living systems and dictate all physiological reactions in conditions of health and disease. In particular, many small proteins play important roles in the regulation of immune responses, inflammation, and neurodegenerative diseases. Therefore, fast and non-invasive protein detection methods can help us create improvements in the areas of disease diagnosis and drug development.
Traditional protein detection methods involve labeling proteins with fluorescent or radioactive markers, to track and detect them. However, this method has proven to be quite expensive, and time consuming. Even more problematic is the fact that these labels can change the function of the protein being studied, making the data collected unreliable. As scientific interest in protein function has increased in recent years, interest in label-free detection methods has also increased. One such method, which is now widely regarded as one of the most effective and sensitive real-time and label-free protein detection techniques, is interferometric scattering microscopy (iSCAT).
iSCAT is based on the sensitive detection of light scattered by individual proteins via interferometry. As each protein settles from the support onto the coverslip, tiny reflections of the protein print on the camera provide information about its size and mass. Therefore, this method is also known as mass photometry. However, the combination of technical noise sources and background fluctuations such as previous spots has limited the sensitivity of iSCAT detection to proteins larger than about 40 KDa.
Using AI to move the microscope boundaries
To push the sensitivity of iSCAT even further, a team from MPL around managing director Vahid Sandoghdar consisting of electrical engineer Mahyar Dahmardeh, computer scientist Houman Mirzaalian, and physical chemist Hisham Mazal have collaborated with Harald Köstler of the Friedrich-Alexander-Universität Erlangen Nürnberg (FAU ) to use two machine learning techniques to detect proteins of only 10 kDa or less.
In a paper published in Nature’s Method (“Self-supervised machine learning pushes the boundaries of sensitivity in the detection of single, unlabeled proteins below 10 kDa”), they show how to use the iForest algorithm in combination with the FastDVDnet technique to achieve this result. Both are called unsupervised machine learning techniques, meaning they don’t need to be trained in advance on labeled data sets. Unsupervised machine learning is highly desirable in microscopy because it allows the identification of patterns and relationships in large data sets without knowing the underlying imaging model.
This is especially important when the detection limit is at the edge of the noise level and there is a shortage of labeled data to train the network.
FastDVDnet is an advanced image denoising technique that removes noise from microscope images using deep neural networks. It is optimized for parallel processing, allowing it to process very large data sets in a relatively short amount of time. In this case, the researchers used FastDVDnet to identify iSCAT protein images from recorded video sequences. The spatiotemporal features extracted by FastDVDnet are then used by iForest to classify iSCAT data.
Unsupervised machine learning algorithms Isolation forest (iForest) is generally used for anomaly detection tasks. It is especially suitable for microscopy because it can handle high-dimensional data with many features, producing more accurate and comprehensive results. This is particularly useful when analyzing microscopy data, where identifying rare or abnormal features is important. For example, iForest anomaly detection can be used to detect the presence of rare structures within biological tissues or to identify cells with unusual morphology. This algorithm can assist in identifying rare or unusual features that may be overlooked by traditional analytical methods.
Professor Vahid Sandoghdar recalls the hard work of his team, but he is also looking forward to the next challenge: “We have come a long way since our first report on the detection of label-free small proteins in Nature Communications in 2014 (“Direct optical sensing of an unlabeled protein and high-resolution imaging of its binding site”). We are determined to push detection boundaries further by improving physical measurement methods and by developing more sophisticated machine learning algorithms. There’s really no underlying reason why we can’t detect molecules below 1kDa, approaching the weight of even a single lipid molecule.”
